Generative artificial intelligence is reshaping transportation data analysis, knowledge service, and decision support, but graduate courses in transportation programs still need a systematic way to connect large model methods with research training. This paper presents a curriculum reform design for Application and Practice of Large Models in Transportation, a 32-hour elective course for first-year master’s students. The reform responds to four problems: fragmented method learning, weak task modeling for multi-source transportation data, insufficient evidence for project-based outputs, and inadequate training in trustworthy use. Based on outcome-based education, the course reconstructs learning objectives, teaching modules, scenario-based projects, and assessment evidence. Prompt design, retrieval-augmented generation, agent-based tool use, experimental evaluation, and academic norms are organized into an integrated pathway. The design emphasizes reproducible project records, data cards, model evaluation cards, and system risk cards. It provides a practical framework for cultivating transportation problem formulation, intelligent application development, research reporting, and trustworthy artificial intelligence awareness.